I was recently (two weeks ago) lucky enough to be invited to
attend an HP Storage Tech Day. It was a
great opportunity to learn some great tech and network with some great people.
I had every intention of getting this post out earlier (honestly!), but the day
job and other engagements have kept me beyond busy. At least I got it out before PEX!
Full disclosures: I work for a VAR where we sell and
implement both HP and EMC storage solutions.
I should also mention that this was an all-expenses paid even. Oh, and we got a really cool jacket with our
twitter handle on the back (which should make me more identifiable at PEX).
The first day started off with an introduction by each of
the attendees. After that, Craig Nunes
got up to give us what was supposed to be the only Marketing part of the
day. He shared with us some charts that
showed the explosive growth HP has seen in the last year, especially around the
3PAR and backup appliance product lines.
He also introduced the HP Storage product line:
- P10000 (3PAR) - Utility platform for ITaaS and Cloud
- P4000 (LeftHand) - Scale out and virtualization
- X#000 (IBRIX) - Scale out NAS and big data
- B#000 (StoreOnce) - Deduplication and disk-based backup
Brad Katz then gave us an overview of the P4000 line,
concentrating on the following:
- VSA - a fully functional virtual appliance version of the
SAN/IQ operating system that runs on the P4000 appliances. It can even be brought into a normal P4000
cluster and clustered across hosts. It's
far more feature rich and scalable than VMware's vSphere Storage Appliance.
- Metrocluster - this is the capability to geographically
separate nodes within a P4000 cluster to create an active-active array across
multiple sites. A very cool
technology. I have customers that run
this capability in their Production environments, and it works great.
- P4800 - by far the coolest technology in the P4000
family. This solution consists of two
BL460 blades that run the VSA and are SAS connected to an MDS disk shelf. The disk shelf is divided into two drawers
(one for each blade) that pull out and contain up to 70TB raw hot swappable
drives. An extremely dense storage
device that utilizes the entire depth of the rack instead of just the front
panel. Check this one in person if you
ever get the chance.
There is also a cool new deployment tool we were able to
play with to deploy the VSA very simply and quickly. They call it "zero to VSA".
Being an iSCSI-only array, some quickly dismiss this product
line for enterprise storage, but I have many happy customers running P4000
storage. Plus I know some NDA futures of
this product line, and there are a lot of cool things to come this year. Some of which should definitely change this
Next, Steve Johnson gave us a briefing of the StoreOnce
products, concentrating on the top of the line B6200. Interesting factoid: HP's first backup
customer was IBM, who OEM'd HP tapes in the 80's.
The B6200 is a combination of DL380s and P2000 hardware and
StoreOnce and IBRIX software. It can
provide both NAS and FC connectivity. It
also provides deduplication using the StoreOnce engine, which is utilized
across all the HP products offering deduplication, a 4k block size and sparse
indexing. It can scale out to 512TB
usable storage and max throughput of 28TB/hr.
Architecturally, it looks a lot like a general purpose SAN
and contains all the redundancy you'd expect (including utilizing RAID6 -
4+2). One major downside: replication is
an add-on license. I find this a major
gap since this array has a strong story as a destination target for multiple
StoreOnce devices in multiple locations.
James "JR" Richardson provided us the 3PAR
presentation. What a great speaker. Very engaging, new his stuff well (he was one
of the first 3PAR employees) and passion to spare. I have every intention of getting him to
Omaha someday soon for a lunch and learn and/or a series of customer meetings.
He spent most of the time laying out the three main
differentiators of the 3PAR line: Multi-tenant, Efficient & Autonomic
For a 3PAR array "multi-tenant" is more than
having multiple customers in a single array.
It's really about handling multiple workloads in a single array, which
is obviously very important in an Infrastructure as a Service model. Virtual Domains do allow for different
subgroups for the creation of storage by different organizations sharing a
The magic in a 3PAR array is in a custom ASIC that handles
all the data movement within the array.
The controller processor handles all the control and data processing,
allowing the ASIC to be built for a single purpose. Each ASIC is also direct
connected (at 2Gbps) to the ASICs on the other controllers, allowing data
movement to be striped across multiple controllers. The ASIC is also built to handle the RAID5
XOR processing, virtually eliminating the write penalty when using RAID5.
3PAR arrays utilize a 16k block size, far smaller than most
arrays, and a 1GB chunklet, allowing for a finer grained autotiering
algorithm. When thin provisioning, there
is now preallocation, so the data written is truly the only data allocated to
the LUN. Snapshots are handled with a redirect on write methodology, which they
claim will allow 100s of snapshots on a single LUN.
Adaptive Optimization (AO) is their autotiering engine. HP has found that 90% of IOPS is in 10% of
the capacity, which is why autotiering is such a hot topic in storage arrays
lately. The most common disk grouping
3PAR arrays contain are RAID5 SSD, RAID1 FC, and RAID6 near-line SAS. I've heard from several 3PAR engineers that
they recommend not initially installing SSD since most RAID1 FC disks can
handle most customer workloads. Within AO, an administrator can define which
tier a write is initially placed in. Policies
can then autotier the block up or down.
Dynamic Optimization (DO) allows for the changing of an
entire LUN or disk group to a new tier.
This is often used to migrate a LUN from RAID1 to RAID5.
Since I have customers who are debating between HP EVA and
3PAR arrays, I asked JR to comment on when one would make sense over the
other. He stated, with very little
hesitation, that he wouldn't buy an EVA at this point. Granted, he did come to HP with the 3PAR
acquisition, but I've never heard any HP employee admit the defeat of the EVA
so directly, especially in such a public setting. He backed his statement up with the fact that
almost all HP Storage R&D is being funneled to the P10000, P4000 and
We were then able to work through a 3PAR lab. While the 3PAR interface wasn't convoluted or
difficult to navigate, they could take some lessons in simplicity from the
P4000 and EVA management interfaces.
We were then taken on a tour of the HP Storage team's lab
environment, including seeing all the products we discussed earlier in the day
and the power/cooling elements of the data center. Some very cool stuff in there, which you can
see in the following pictures.
We finished up the presentations for the day with the
X-series of products, the provide file-based storage on the IBRIX platform.
Jim Hankins, product manager for IBRIX, and Chris Duffy,
product manager for file storage, kicked off the presentation with an overview
of the product line. This includes the
X1000 (integrated), X3000 (gateway), and X5000 (integrated), all of which are
Windows Storage Server based and include file and block storage, deduplication,
snapshots, and file classification
Mark Tomas, IBRIX development manager, then covered the
X9000, which is designed for archiving and massive scale out. The top of the line is the X9720, which can
contain up to 1.2PB using 2TB MDL-SAS disks.
A few of the features include:
- Uses MDS shelves and blades (similar to P4800)
- Single namespace across all models allowing tiering
- A single X9000 namespace can grow up to 16PB
- Tiering is based on metadata
- Continuous remote replication is available to a node in
the same cluster or another cluster
Robert Thompson, an Architect for Windows NAS Engineering,
presented the X5000, a converged network storage server. This is very cool technology and is the best
example of converged infrastructure I've ever seen. Imagine a 2U device that contains a couple of
BL460s and a small drawer of disks containing 16LFF disks. The blades run a Windows Storage Server
cluster and direct attach to the disks in the drawer. Up to 4 HP D2000 disk shelves can be direct
attached to the device allowing for a maximum of 128TB with 116 disks (16LFF +
100SFF). The possibilities for this
hardware platform are very interesting (an E5000 Exchange appliance has already
been released) and HP is currently listening very closely to customer feedback,
so let them know your thoughts (think branch office vSphere cluster).
The second day started with a VMware integration
presentation by fellow vExpert Eric Siebert and Aboubacar Diare (another guy
I'd like to get to Omaha for some presentations). Currently, the only gaps in VMware
integration within the current HP Storage portfolio are VAAI in the P6000 (in
QA right now), VASA on the P2000, and SRM 4.1 & 5.0 support for P2000. If you want to use VASA with your HP storage,
it requires the Insight Control Storage module or 3PAR Management Plug-in,
which both act as the VASA provider.
They will then provide vCenter the array type, LUN type, snapshot
information, whether the LUN is thin provisioned, CA relationship, and replication
They discussed how, due to Zero detect and Write Same
features in the 3PAR, Eager Zero Thick (EZT) VMDKs on a 3PAR LUN can actually
be faster to create (definitely not slower) than Lazy Zero Thick (LZT)
VMDKs. Of course, EZT VMDKs will provide
better performance on writes to the VMDK once created. For XCOPY & Write Same operations, the
3PAR, EVA, P9500 platforms all have abilities to throttle VAAI operations so as
to not harm front end IO.
We finished up the presentations with Mike Koponen who
presented the BladeSystem product line, which are designed for a quick and easy
implementation of the hardware for a vSphere, XenServer or Hyper-V
environment. The line includes the
- VS1 - Modular scalability, built on DL360s and P4500 and
supporting VMware or Microsoft. Both the
servers & P4500 start at 2 nodes, and both can scale out to 8 nodes
each. The vCenter server and the P4000
Centralized Management Console are both virtualized on their own DL360.
- VS2 - Density & flexibility, built on BL460s and P4800
and supporting VMware or Microsoft. It
starts with 6 blades, but can scale to 12.
The P4800 can contain one node (42TB) or two nodes (84TB). It is also upgradable to CloudSystem.
- CV2 - Similar architecture to the VS2, but offering both
SAN & DAS versions, is designed specifically for client virtualization, and
supports VMware or Citrix environments.
- VS3 - Scale & performance, built on BL460s and 3PAR
and only supports VMware. It starts out
with 16 blades and is scalable to 64.
The storage is built with a F-series 3PAR and starts at 38.4TB and can
be grown to 162.2TB. It is also
upgradable to CloudSystem.
All four models include base and extended SKUs.
Overall, a great experience.
I learned a bunch, and hopefully this post will help you out as well.
A big thank you to HP for putting this together and inviting
me to join!
For additional coverage of the conference, check out Calvin Zito's blog for his posts and links to the other attendee's posts.